we are using 0805 version of tidl. use the latest scatterND layer.
The current phenomenon is that graph verify said： cloud not allocate tensor memory.
After changing the memory allocation, it can be determined that there is enough memory on c7x scratch, about 800MB.
After removing scatterND, we don't see an error.
The difference is two inputs to sactterND，the first intput shape is: [64, 304, 496]，the second input shape is: [10000, 1, 10]
Is there any ideas to check this error?
